Target portrait screening identification method and system based on relative similarity weighted average, and readable medium

Abstract

The invention belongs to the technical field of artificial intelligence image processing, and relates to a relative similarity weighted average-based target portrait screening recognition method and system and a readable medium, and the method comprises the following steps: selecting a plurality of different types of faces of a target person; inputting the different types of human faces and each human face in the candidate library into a universal human face recognition model for genetic relationship recognition, and outputting the similarity between the different types of human faces and the human faces in the candidate library; and performing weighted average on each similarity to obtain a final target portrait screening recognition result. The method mainly solves the problem that a current genetic relationship face recognition algorithm is low in accuracy and poor in robustness, so that a genetic relationship portrait comparison system which can be actually used is lacked.

Description

Technical Field

[0001] The present invention relates to a method, system and readable medium for screening and identifying target portraits based on weighted average of kinship similarity, and belongs to the technical field of artificial intelligence image processing. Background Art

[0002] The kinship portrait matching technology is a method that uses biometric recognition technology to determine the kinship relationship between individuals. This technology is mainly used in the fields of forensic medicine, population genetics and social security, especially in finding long-lost relatives, criminal investigation and parent-child determination.

[0003] Traditional kinship portrait comparison techniques are usually based on facial recognition algorithms or methods, analyzing the facial features of an individual and comparing them with those of his or her relatives. These comparisons rely on the more heritable parts of facial features, such as the shape and position of the eyes, nose, and mouth. Currently, AI-based portrait comparison techniques are also beginning to be applied to kinship portrait comparisons. By training models on facial image data of kinship portraits, a high accuracy rate has been achieved in small-scale research data.

[0004] However, whether it is the traditional kinship portrait algorithms and methods, or the current AI-based portrait comparison technology, their accuracy in real scenes is relatively low, making it difficult to move beyond the laboratory and achieve large-scale accurate applications. Summary of the invention

[0005] In view of the above problems, the purpose of the present invention is to provide a method, system and readable medium for screening and identifying target portraits based on weighted average of kinship similarity, mainly to solve the problem that the current kinship face recognition algorithm has low accuracy and poor robustness, resulting in a lack of a kinship portrait comparison system that can be used in practice.

[0006] To achieve the above-mentioned purpose, the present invention proposes the following technical scheme: a method for screening and identifying target portraits based on weighted averaging of kinship similarities, comprising the following steps: selecting multiple different types of faces of a target person; inputting each of the different types of faces and each face in a candidate library into a universal face recognition model for kinship recognition, and outputting the similarities between the different types of faces and the faces in the candidate library; and obtaining the final target portrait screening and recognition result after weighted averaging each of the similarities.

[0007] Further, different types of faces include faces of related persons and faces of unrelated persons.

[0008] Furthermore, the number of the candidate libraries is greater than or equal to two. Furthermore, N types of faces are compared with each face in the candidate library, and similarity scores are obtained. The score with the highest similarity in each type is selected as the similarity score of the type. Each candidate library obtains N different types of similarity scores. According to the weight of each type, a weighted average similarity is obtained. The weighted averages of each candidate library are sorted from high to low to obtain the target portrait screening and recognition results.

[0009] Furthermore, the calculation formula of the weighted average similarity is:

[0010] in, is the weighted average similarity, is the similarity of the first category, is the similarity of the second category, is the similarity of the Nth class, is the weight of the first class, is the weight of the second category, is the weight of the Nth class.

[0011] Furthermore, the general face recognition model includes a data preprocessing module, a feature extraction module, a feature vector generation module and a similarity calculation module. The data preprocessing module is used to detect the face area in the input image, crop the face part, and align the face; the feature extraction module is used to identify the feature points in the face, perform feature extraction, and generate a feature map; the feature vector generation module is used to flatten the feature map through a fully connected layer and convert it into a vector output of a fixed size; the similarity calculation module is used to calculate the similarity between the input face and the face feature vectors in the candidate library.

[0012] Furthermore, the general face recognition model includes an open source comparison model or a closed source comparison model; the open source comparison model includes Face Recognition, InsightFace and OpenFace software models; the closed source comparison model includes software models of SenseTime, Yitu, Yuncong and Green Deepin.

[0013] Furthermore, the more different types of faces there are, the stronger the kinship determination capability of the universal face recognition model is.

[0014] The present invention also discloses a target portrait screening and recognition system based on weighted average of kinship similarity, comprising: a face selection module, used to select multiple different types of faces of the target person; a similarity acquisition module, used to input each of the different types of faces and faces in a candidate library into a universal face recognition model for kinship recognition, and output the similarity between the different types of faces and faces in the candidate library; a portrait screening module, used to obtain the final target portrait screening and recognition result after weighted average of each of the similarities.

[0015] The present invention also discloses a computer-readable storage medium, on which a computer program is stored. The computer program is executed by a processor to implement any of the above-mentioned target portrait screening and recognition methods based on weighted average of kinship similarity.

[0016] The technical solution of the present invention has at least the following technical effects or advantages: 1. The solution of the patent of the present invention reduces the reliance on the recognition accuracy of a single family member by comparing the photos of multiple relatives with all the people in the candidate library and taking a weighted average of the similarity scores. This method can be more robust when dealing with complex family structures containing extensive kinship relationships because it does not directly rely on direct comparisons between specific family members.

[0017] 2. The present invention uses a universal face recognition model, which does not require a large amount of individual information for training, thus avoiding the problem of data privacy.

[0018] 3. Since the focus is on statistical analysis of similarity rather than complex facial feature prediction models, the solution of the present invention may show better generalization ability in different populations and ethnic backgrounds. This statistical method-based processing reduces the risk of overfitting and makes the algorithm more suitable for diverse application environments.

[0019] 4. The solution of the present invention may be more efficient in processing large amounts of data by simplifying the data processing steps.

[0020] 5. Compared with complex model training and specific parameter settings, the solution of the present invention focuses on a more universal statistical method, which enables it to adapt to different technical and application requirements more flexibly. In addition, the parameter adjustment of this method is relatively simple, which is convenient for rapid adjustment and optimization in different application scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 is a flow chart of a method for screening and identifying target portraits based on weighted average of kinship similarity in one embodiment of the present invention; Figure 2 is a graph showing the test results of the ability of a portrait identity recognition algorithm to recognize related portraits in one embodiment of the present invention; Figure 3 In one embodiment of the present invention, (a) is a similarity score distribution diagram of the comparison results when N=3, that is, three related portraits are used simultaneously to compare with the portrait to be compared; (b) is a similarity score distribution diagram when N=5, that is, five related portraits are used simultaneously to compare with the portrait to be compared. DETAILED DESCRIPTION

[0022] In order to enable those skilled in the art to better understand the technical solution of the present invention, the present invention is described in detail through specific embodiments. However, it should be understood that the provision of specific embodiments is only for a better understanding of the present invention, and they should not be understood as limitations of the present invention. In the description of the present invention, it should be understood that the terms used are only for the purpose of description and cannot be understood as indicating or implying relative importance.

[0023] In order to solve the problems existing in the existing technology, whether it is the traditional kinship portrait algorithms and methods, or the current portrait comparison technology based on artificial intelligence, its accuracy in real scenes is relatively low, making it difficult to go out of the laboratory; and to achieve large-scale accurate applications. The reasons for the low accuracy of kinship portrait comparison technology also include the following: Genetic diversity: The diversity of the human genome means that even closely related individuals may have very different facial features, which increases the difficulty of comparing portraits of closely related individuals.

[0024] Influence of environmental factors: External factors such as an individual's growth environment and living habits can also affect facial features, reducing the similarity between related individuals.

[0025] Age differences: The facial features of individuals at different ages will change significantly, which further increases the difficulty of comparing portraits of relatives across age groups.

[0026] Insufficient data: The data sets currently available for kinship portrait comparison studies are relatively small, making it difficult to train a model that is robust and has good generalization capabilities.

[0027] The present invention proposes a method, system and readable medium for screening and identifying target portraits based on weighted average of kinship similarity. On the basis of using a portrait identity recognition algorithm, combined with genetic knowledge and big data analysis technology, a relatively accurate kinship portrait comparison system is realized to improve the accuracy and practicality of kinship portrait comparison. The scheme of the present invention is described in detail below through embodiments in conjunction with the accompanying drawings.

[0028] Embodiment 1 This embodiment discloses a method for screening and identifying target portraits based on weighted average of kinship similarity, comprising the following steps: S1 selects multiple faces of different types of the target person; S2 inputs different types of faces and faces in the candidate library into a universal face recognition model for kinship recognition, and outputs the similarity between different types of faces and faces in the candidate library; The universal face recognition model of this embodiment includes a data preprocessing module, a feature extraction module, a feature vector generation module and a similarity calculation module. The data preprocessing module is used to detect the face area in the input image, crop the face part, and align the face; The feature extraction module is used to identify the feature points in the face, extract features, and generate feature maps. The feature points can be eyes, noses, corners of the mouth, etc. The feature extraction module includes a convolution layer, an activation function layer, and a pooling layer. The convolution layer is used to extract the feature maps of the image from low-level to high-level. In this embodiment, the activation function selects the ReLU function to introduce nonlinear features; the pooling layer is used to reduce the size of the feature map, retain important features, and reduce the amount of calculation.

[0029] The feature vector generation module is used to flatten the feature map through the fully connected layer and convert it into a vector output of a fixed size; The similarity calculation module is used to calculate the similarity between the input face and the face feature vectors in the candidate library.

[0030] General face recognition models include open source comparison models or closed source comparison models; open source comparison models include FaceRecognition, InsightFace and OpenFace software models; closed source comparison models include software models of SenseTime, Yitu, Yuncong and Green Deepin. The models used in this embodiment are not limited to the models mentioned above.

[0031] First, we tested its ability to verify kinship using large-scale data, using a large number of portrait groups, with different types of faces including those with kinship and those without kinship. Figure 2 As shown in the figure, the dark area is the algorithm's score for the similarity of data with kinship, and the light area is the algorithm's score for the similarity of data without kinship. The overall score for the portraits of people with kinship is slightly higher than that of people without kinship. It can be seen that the overlapping area of ​​the two is large, so it is difficult to determine kinship through this score. Considering that there are multiple portraits of relatives that can be obtained in the process of kinship determination (set as N), we compare these N relatives with the person to be compared, obtain N similarity scores, and then average these N similarity scores. When N=3, 5, a new similarity score distribution diagram can be calculated, as shown below: Figure 3 shown.

[0032] S3 performs weighted averaging on each similarity to obtain the final target portrait screening and recognition result.

[0033] In this embodiment, the number of candidate libraries is greater than or equal to two, and N types of faces are compared with each face in the candidate library to obtain a similarity score. The highest similarity score in each type is selected as the similarity score of the type. Each candidate library obtains N different types of similarity scores. According to the weight of each type, a weighted average similarity is obtained. The weighted average of each candidate library is sorted from high to low to obtain the target portrait screening and recognition result. The calculation formula for weighted average similarity is:

[0034] in, is the weighted average similarity, is the similarity of the first category, is the similarity of the second category, is the similarity of the Nth class, is the weight of the first class, is the weight of the second category, is the weight of the Nth class.

[0035] In this embodiment, the initial value of each type of weight is set to 1 / N, and the weight of each type is adjusted as the model is trained. The value of the weight can also be set manually.

[0036] It can be seen that when N is larger, the overlapping area of ​​the scores for having / not having a kinship relationship becomes smaller and smaller, thereby achieving a stronger ability to determine kinship relationships. By comparing all the portraits in the database to be compared with these N relatives and then sorting the results, it is possible to more accurately identify targets with kinship relationships.

[0037] This embodiment overcomes the shortcomings of the prior art in that it is inefficient and inaccurate to manually extract and calculate features, as well as the shortcomings of being difficult to train and implement due to lack of data, and realizes a new method for comparing kinship portraits with strong robustness and robustness.

[0038] Embodiment 2 Based on the same inventive concept, this embodiment discloses a target portrait screening and recognition system based on weighted average of kinship similarity, comprising: A face selection module is used to select multiple faces of different types of target persons; A similarity acquisition module is used to input different types of faces and faces in the candidate library into a universal face recognition model for kinship recognition, and output the similarity between different types of faces and faces in the candidate library; The portrait screening module is used to obtain the final target portrait screening and recognition result after weighted averaging of each similarity.

[0039] Embodiment 3 Based on the same inventive concept, this embodiment discloses a computer-readable storage medium, on which a computer program is stored. The computer program is executed by a processor to implement any of the above-mentioned target portrait screening and recognition methods based on weighted average of kinship similarity.

[0040] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0041] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0042] These computer program instructions may also be stored in a computer readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0043] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0044] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit them. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation of the present invention can still be modified or replaced by equivalents, and any modification or equivalent replacement that does not deviate from the spirit and scope of the present invention should be included in the protection scope of the claims of the present invention. The above content is only a specific implementation of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of changes or replacements within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A target portrait screening and recognition method based on weighted average of kinship similarity, It is characterized in that The following steps are involved: Select multiple faces of different types of the target person; Inputting each of the different types of faces and each face in the candidate library into a universal face recognition model for kinship recognition, and outputting the similarity between the different types of faces and the faces in the candidate library; After weighted averaging of the similarities, the final target portrait screening and recognition result is obtained.

2. The target portrait screening and recognition method based on weighted average of kinship similarity as claimed in claim 1, It is characterized in that Different types of faces include faces of related persons and faces of unrelated persons.

3. The target portrait screening and recognition method based on weighted average of kinship similarity as claimed in claim 1, It is characterized in that The number of the candidate libraries is greater than or equal to two.

4. The target portrait screening and recognition method based on weighted average of kinship similarity as claimed in claim 3, It is characterized in that N types of faces are compared with each face in the candidate library, and similarity scores are obtained. The score with the highest similarity in each type is selected as the similarity score of the type. Each candidate library obtains N different types of similarity scores. According to the weight of each type, a weighted average similarity is obtained. The weighted averages of each candidate library are sorted from high to low to obtain the target portrait screening and recognition results.

5. The target person portrait screening and recognition method based on weighted average of kinship similarity as claimed in claim 4, It is characterized in that The calculation formula of the weighted average similarity is: in, is the weighted average similarity, is the similarity of the first category, is the similarity of the second category, is the similarity of the Nth class, is the weight of the first class, is the weight of the second category, is the weight of the Nth class.

6. The target portrait screening and recognition method based on weighted average of kinship similarity according to any one of claims 1 to 5, It is characterized in that The general face recognition model includes a data preprocessing module, a feature extraction module, a feature vector generation module and a similarity calculation module. The data preprocessing module is used to detect the face area in the input image, crop the face part, and align the face; The feature extraction module is used to identify feature points in the face, perform feature extraction, and generate a feature map; The feature vector generation module is used to flatten the feature map through a fully connected layer and convert it into a vector output of a fixed size; The similarity calculation module is used to calculate the similarity between the input face and the face feature vectors in the candidate library.

7. The target portrait screening and recognition method based on weighted average of kinship similarity as claimed in claim 6, It is characterized in that The general face recognition model includes an open source comparison model or a closed source comparison model; the open source comparison model includes Face Recognition, InsightFace and OpenFace software models; the closed source comparison model includes software models of SenseTime, Yitu, Yuncong and Green Deepin.

8. The target portrait screening and recognition method based on weighted average of kinship similarity according to any one of claims 1 to 5, It is characterized in that The more different types of faces there are, the stronger the kinship determination capability of the universal face recognition model is.

9. A target portrait screening and recognition system based on weighted average of kinship similarity, It is characterized in that include: A face selection module is used to select multiple faces of different types of target persons; A similarity acquisition module, used for inputting each of the different types of faces and faces in the candidate library into a universal face recognition model for kinship recognition, and outputting the similarity between the different types of faces and faces in the candidate library; The portrait screening module is used to obtain the final target portrait screening and recognition result after weighted averaging the similarities.

10. A computer-readable storage medium, It is characterized in that The computer-readable storage medium stores a computer program, and the computer program is executed by a processor to implement the target portrait screening and recognition method based on weighted average of kinship similarity as described in any one of claims 1 to 8.

Images